drivers/base/cpu.c

   1 /*
   2  * CPU subsystem support
   3  */
   4
   5 #include <linux/kernel.h>
   6 #include <linux/module.h>
   7 #include <linux/init.h>
   8 #include <linux/sched.h>
   9 #include <linux/cpu.h>
  10 #include <linux/topology.h>
  11 #include <linux/device.h>
  12 #include <linux/node.h>
  13 #include <linux/gfp.h>
  14 #include <linux/slab.h>
  15 #include <linux/percpu.h>
  16 #include <linux/cpufeature.h>
  17
  18 #include "base.h"
  19
  20 struct bus_type cpu_subsys = {
  21         .name = "cpu",
  22         .dev_name = "cpu",
  23 };
  24 EXPORT_SYMBOL_GPL(cpu_subsys);
  25
  26 static DEFINE_PER_CPU(struct device *, cpu_sys_devices);
  27
  28 #ifdef CONFIG_HOTPLUG_CPU
  29 static void change_cpu_under_node(struct cpu *cpu,
  30                         unsigned int from_nid, unsigned int to_nid)
  31 {
  32         int cpuid = cpu->dev.id;
  33         unregister_cpu_under_node(cpuid, from_nid);
  34         register_cpu_under_node(cpuid, to_nid);
  35         cpu->node_id = to_nid;
  36 }
  37
  38 static ssize_t show_online(struct device *dev,
  39                            struct device_attribute *attr,
  40                            char *buf)
  41 {
  42         struct cpu *cpu = container_of(dev, struct cpu, dev);
  43
  44         return sprintf(buf, "%u\n", !!cpu_online(cpu->dev.id));
  45 }
  46
  47 static ssize_t __ref store_online(struct device *dev,
  48                                   struct device_attribute *attr,
  49                                   const char *buf, size_t count)
  50 {
  51         struct cpu *cpu = container_of(dev, struct cpu, dev);
  52         int cpuid = cpu->dev.id;
  53         int from_nid, to_nid;
  54         ssize_t ret;
  55
  56         cpu_hotplug_driver_lock();
  57         switch (buf[0]) {
  58         case '0':
  59                 ret = cpu_down(cpuid);
  60                 if (!ret)
  61                         kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
  62                 break;
  63         case '1':
  64                 from_nid = cpu_to_node(cpuid);
  65                 ret = cpu_up(cpuid);
  66
  67                 /*
  68                  * When hot adding memory to memoryless node and enabling a cpu
  69                  * on the node, node number of the cpu may internally change.
  70                  */
  71                 to_nid = cpu_to_node(cpuid);
  72                 if (from_nid != to_nid)
  73                         change_cpu_under_node(cpu, from_nid, to_nid);
  74
  75                 if (!ret)
  76                         kobject_uevent(&dev->kobj, KOBJ_ONLINE);
  77                 break;
  78         default:
  79                 ret = -EINVAL;
  80         }
  81         cpu_hotplug_driver_unlock();
  82
  83         if (ret >= 0)
  84                 ret = count;
  85         return ret;
  86 }
  87 static DEVICE_ATTR(online, 0644, show_online, store_online);
  88
  89 static void __cpuinit register_cpu_control(struct cpu *cpu)
  90 {
  91         device_create_file(&cpu->dev, &dev_attr_online);
  92 }
  93 void unregister_cpu(struct cpu *cpu)
  94 {
  95         int logical_cpu = cpu->dev.id;
  96
  97         unregister_cpu_under_node(logical_cpu, cpu_to_node(logical_cpu));
  98
  99         device_remove_file(&cpu->dev, &dev_attr_online);
 100
 101         device_unregister(&cpu->dev);
 102         per_cpu(cpu_sys_devices, logical_cpu) = NULL;
 103         return;
 104 }
 105
 106 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
 107 static ssize_t cpu_probe_store(struct device *dev,
 108                                struct device_attribute *attr,
 109                                const char *buf,
 110                                size_t count)
 111 {
 112         return arch_cpu_probe(buf, count);
 113 }
 114
 115 static ssize_t cpu_release_store(struct device *dev,
 116                                  struct device_attribute *attr,
 117                                  const char *buf,
 118                                  size_t count)
 119 {
 120         return arch_cpu_release(buf, count);
 121 }
 122
 123 static DEVICE_ATTR(probe, S_IWUSR, NULL, cpu_probe_store);
 124 static DEVICE_ATTR(release, S_IWUSR, NULL, cpu_release_store);
 125 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
 126
 127 #else /* ... !CONFIG_HOTPLUG_CPU */
 128 static inline void register_cpu_control(struct cpu *cpu)
 129 {
 130 }
 131 #endif /* CONFIG_HOTPLUG_CPU */
 132
 133 #ifdef CONFIG_KEXEC
 134 #include <linux/kexec.h>
 135
 136 static ssize_t show_crash_notes(struct device *dev, struct device_attribute *attr,
 137                                 char *buf)
 138 {
 139         struct cpu *cpu = container_of(dev, struct cpu, dev);
 140         ssize_t rc;
 141         unsigned long long addr;
 142         int cpunum;
 143
 144         cpunum = cpu->dev.id;
 145
 146         /*
 147          * Might be reading other cpu's data based on which cpu read thread
 148          * has been scheduled. But cpu data (memory) is allocated once during
 149          * boot up and this data does not change there after. Hence this
 150          * operation should be safe. No locking required.
 151          */
 152         addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpunum));
 153         rc = sprintf(buf, "%Lx\n", addr);
 154         return rc;
 155 }
 156 static DEVICE_ATTR(crash_notes, 0400, show_crash_notes, NULL);
 157
 158 static ssize_t show_crash_notes_size(struct device *dev,
 159                                      struct device_attribute *attr,
 160                                      char *buf)
 161 {
 162         ssize_t rc;
 163
 164         rc = sprintf(buf, "%zu\n", sizeof(note_buf_t));
 165         return rc;
 166 }
 167 static DEVICE_ATTR(crash_notes_size, 0400, show_crash_notes_size, NULL);
 168 #endif
 169
 170 /*
 171  * Print cpu online, possible, present, and system maps
 172  */
 173
 174 struct cpu_attr {
 175         struct device_attribute attr;
 176         const struct cpumask *const * const map;
 177 };
 178
 179 static ssize_t show_cpus_attr(struct device *dev,
 180                               struct device_attribute *attr,
 181                               char *buf)
 182 {
 183         struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr);
 184         int n = cpulist_scnprintf(buf, PAGE_SIZE-2, *(ca->map));
 185
 186         buf[n++] = '\n';
 187         buf[n] = '\0';
 188         return n;
 189 }
 190
 191 #define _CPU_ATTR(name, map) \
 192         { __ATTR(name, 0444, show_cpus_attr, NULL), map }
 193
 194 /* Keep in sync with cpu_subsys_attrs */
 195 static struct cpu_attr cpu_attrs[] = {
 196         _CPU_ATTR(online, &cpu_online_mask),
 197         _CPU_ATTR(possible, &cpu_possible_mask),
 198         _CPU_ATTR(present, &cpu_present_mask),
 199 };
 200
 201 /*
 202  * Print values for NR_CPUS and offlined cpus
 203  */
 204 static ssize_t print_cpus_kernel_max(struct device *dev,
 205                                      struct device_attribute *attr, char *buf)
 206 {
 207         int n = snprintf(buf, PAGE_SIZE-2, "%d\n", NR_CPUS - 1);
 208         return n;
 209 }
 210 static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL);
 211
 212 /* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */
 213 unsigned int total_cpus;
 214
 215 static ssize_t print_cpus_offline(struct device *dev,
 216                                   struct device_attribute *attr, char *buf)
 217 {
 218         int n = 0, len = PAGE_SIZE-2;
 219         cpumask_var_t offline;
 220
 221         /* display offline cpus < nr_cpu_ids */
 222         if (!alloc_cpumask_var(&offline, GFP_KERNEL))
 223                 return -ENOMEM;
 224         cpumask_andnot(offline, cpu_possible_mask, cpu_online_mask);
 225         n = cpulist_scnprintf(buf, len, offline);
 226         free_cpumask_var(offline);
 227
 228         /* display offline cpus >= nr_cpu_ids */
 229         if (total_cpus && nr_cpu_ids < total_cpus) {
 230                 if (n && n < len)
 231                         buf[n++] = ',';
 232
 233                 if (nr_cpu_ids == total_cpus-1)
 234                         n += snprintf(&buf[n], len - n, "%d", nr_cpu_ids);
 235                 else
 236                         n += snprintf(&buf[n], len - n, "%d-%d",
 237                                                       nr_cpu_ids, total_cpus-1);
 238         }
 239
 240         n += snprintf(&buf[n], len - n, "\n");
 241         return n;
 242 }
 243 static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL);
 244
 245 static void cpu_device_release(struct device *dev)
 246 {
 247         /*
 248          * This is an empty function to prevent the driver core from spitting a
 249          * warning at us.  Yes, I know this is directly opposite of what the
 250          * documentation for the driver core and kobjects say, and the author
 251          * of this code has already been publically ridiculed for doing
 252          * something as foolish as this.  However, at this point in time, it is
 253          * the only way to handle the issue of statically allocated cpu
 254          * devices.  The different architectures will have their cpu device
 255          * code reworked to properly handle this in the near future, so this
 256          * function will then be changed to correctly free up the memory held
 257          * by the cpu device.
 258          *
 259          * Never copy this way of doing things, or you too will be made fun of
 260          * on the linux-kernel list, you have been warned.
 261          */
 262 }
 263
 264 #ifdef CONFIG_HAVE_CPU_AUTOPROBE
 265 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE
 266 static ssize_t print_cpu_modalias(struct device *dev,
 267                                   struct device_attribute *attr,
 268                                   char *buf)
 269 {
 270         ssize_t n;
 271         u32 i;
 272
 273         n = sprintf(buf, "cpu:type:" CPU_FEATURE_TYPEFMT ":feature:",
 274                     CPU_FEATURE_TYPEVAL);
 275
 276         for (i = 0; i < MAX_CPU_FEATURES; i++)
 277                 if (cpu_have_feature(i)) {
 278                         if (PAGE_SIZE < n + sizeof(",XXXX\n")) {
 279                                 WARN(1, "CPU features overflow page\n");
 280                                 break;
 281                         }
 282                         n += sprintf(&buf[n], ",%04X", i);
 283                 }
 284         buf[n++] = '\n';
 285         return n;
 286 }
 287 #else
 288 #define print_cpu_modalias      arch_print_cpu_modalias
 289 #endif
 290
 291 static int cpu_uevent(struct device *dev, struct kobj_uevent_env *env)
 292 {
 293         char *buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
 294         if (buf) {
 295                 print_cpu_modalias(NULL, NULL, buf);
 296                 add_uevent_var(env, "MODALIAS=%s", buf);
 297                 kfree(buf);
 298         }
 299         return 0;
 300 }
 301 #endif
 302
 303 /*
 304  * register_cpu - Setup a sysfs device for a CPU.
 305  * @cpu - cpu->hotpluggable field set to 1 will generate a control file in
 306  *        sysfs for this CPU.
 307  * @num - CPU number to use when creating the device.
 308  *
 309  * Initialize and register the CPU device.
 310  */
 311 int __cpuinit register_cpu(struct cpu *cpu, int num)
 312 {
 313         int error;
 314
 315         cpu->node_id = cpu_to_node(num);
 316         memset(&cpu->dev, 0x00, sizeof(struct device));
 317         cpu->dev.id = num;
 318         cpu->dev.bus = &cpu_subsys;
 319         cpu->dev.release = cpu_device_release;
 320 #ifdef CONFIG_ARCH_HAS_CPU_AUTOPROBE
 321         cpu->dev.bus->uevent = cpu_uevent;
 322 #endif
 323         error = device_register(&cpu->dev);
 324         if (!error && cpu->hotpluggable)
 325                 register_cpu_control(cpu);
 326         if (!error)
 327                 per_cpu(cpu_sys_devices, num) = &cpu->dev;
 328         if (!error)
 329                 register_cpu_under_node(num, cpu_to_node(num));
 330
 331 #ifdef CONFIG_KEXEC
 332         if (!error)
 333                 error = device_create_file(&cpu->dev, &dev_attr_crash_notes);
 334         if (!error)
 335                 error = device_create_file(&cpu->dev,
 336                                            &dev_attr_crash_notes_size);
 337 #endif
 338         return error;
 339 }
 340
 341 struct device *get_cpu_device(unsigned cpu)
 342 {
 343         if (cpu < nr_cpu_ids && cpu_possible(cpu))
 344                 return per_cpu(cpu_sys_devices, cpu);
 345         else
 346                 return NULL;
 347 }
 348 EXPORT_SYMBOL_GPL(get_cpu_device);
 349
 350 #ifdef CONFIG_HAVE_CPU_AUTOPROBE
 351 static DEVICE_ATTR(modalias, 0444, print_cpu_modalias, NULL);
 352 #endif
 353
 354 static struct attribute *cpu_root_attrs[] = {
 355 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
 356         &dev_attr_probe.attr,
 357         &dev_attr_release.attr,
 358 #endif
 359         &cpu_attrs[0].attr.attr,
 360         &cpu_attrs[1].attr.attr,
 361         &cpu_attrs[2].attr.attr,
 362         &dev_attr_kernel_max.attr,
 363         &dev_attr_offline.attr,
 364 #ifdef CONFIG_HAVE_CPU_AUTOPROBE
 365         &dev_attr_modalias.attr,
 366 #endif
 367         NULL
 368 };
 369
 370 static struct attribute_group cpu_root_attr_group = {
 371         .attrs = cpu_root_attrs,
 372 };
 373
 374 static const struct attribute_group *cpu_root_attr_groups[] = {
 375         &cpu_root_attr_group,
 376         NULL,
 377 };
 378
 379 bool cpu_is_hotpluggable(unsigned cpu)
 380 {
 381         struct device *dev = get_cpu_device(cpu);
 382         return dev && container_of(dev, struct cpu, dev)->hotpluggable;
 383 }
 384 EXPORT_SYMBOL_GPL(cpu_is_hotpluggable);
 385
 386 #ifdef CONFIG_GENERIC_CPU_DEVICES
 387 static DEFINE_PER_CPU(struct cpu, cpu_devices);
 388 #endif
 389
 390 static void __init cpu_dev_register_generic(void)
 391 {
 392 #ifdef CONFIG_GENERIC_CPU_DEVICES
 393         int i;
 394
 395         for_each_possible_cpu(i) {
 396                 if (register_cpu(&per_cpu(cpu_devices, i), i))
 397                         panic("Failed to register CPU device");
 398         }
 399 #endif
 400 }
 401
 402 void __init cpu_dev_init(void)
 403 {
 404         if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups))
 405                 panic("Failed to register CPU subsystem");
 406
 407         cpu_dev_register_generic();
 408 }